Harmonic Morphisms with Fibers of Dimension One
Total Page:16
File Type:pdf, Size:1020Kb
COMMUNICATIONS IN ANALYSIS AND GEOMETRY Volume 8, Number 2, 219-265, 2000 Harmonic morphisms with fibers of dimension one ROBERT L. BRYANT1 The harmonic morphisms (j) : Mn+1 —> Nn are studied using the methods of the moving frame and exterior differential systems and three main results are achieved. The first result is a local structure theorem for such maps in the case that (f) is a submersion, in particular, a normal form is found for all such (j) once the metric on the target manifold N is specified. The second result is a finiteness theorem, which says, in a certain sense, that, when n > 3, the set of harmonic morphisms with a given Riemannian domain (Mn+1, #) is a finite dimensional space. The third result is the explicit classification when n > 3 of all local and global harmonic morphisms with domain (Mn+1,p), a space of constant curvature. 0. Introduction. A smooth map <j) : M —> N between Riemannian manifolds is said to be a harmonic morphism if, for any harmonic function / on any open set V C iV, the pullback / o </> is a harmonic function on (^^(V) C M. By a simple argument (see §1), any non-constant harmonic morphism </>: M —> N between connected Riemannian manifolds must be a submersion away from a set of measure zero in M. Thus, a necessary condition for the existence of a non-constant harmonic map 0 : M —* N is that dimM > dimN. When the dimension of N is 1, so that iV can be regarded, at least locally, as R with its standard metric, a map </> : M -* N is a harmonic morphism if and only if it is a harmonic function in the usual sense. Thus, at least locally, there are many harmonic morphisms from M to N. 1This research was begun during a visit to IMPA in Rio de Janeiro in July 1996 and was inspired by questions raised during the International Conference on Differential Geometry held at IMPA during that month. The article was written during a visit to the Institute for Advanced Study in Princeton. The author would like to thank IMPA and the IAS for their hospitality and also to acknowledge support from the National Science Foundation through grant DMS-9505125. 219 220 Robert Bryant However, when the dimension of N is greater than 1, the condition of being a harmonic morphism turns out to be much more restrictive, being essentially equivalent to an overdetermined system of PDE for the map </>. Thus, for generic Riemannian metrics on M and AT, one does not expect there to be any harmonic morphisms, even locally. Moreover, in the case that there do exist harmonic morphisms (j) : M —> N for given M and AT, one expects the analysis of the overdetermined system that describes them to involve integrability conditions and other features of overdetermined sys- tems. When both M and N have dimension 2, a harmonic morphism is sim- ply a branched conformal mapping between Riemann surfaces and these are studied by classical methods of complex analysis and Riemann surface theory. When both manifolds have the same dimension n > 2, a non-constant harmonic morphism is a local homothety, i.e., up to a constant scale factor, (j) is a local isometry. Thus, the interesting cases are when dimM > dim AT > 2. This article concerns the case when dimM = dimN + 1, i.e., when the dimension of the generic fiber of (/> is 1. It contains three main results. The first, Theorem 1, is a local structure theorem for harmonic mor- phisms whose fibers are curves. This result describes the possible Rieman- nian metrics g that can be defined on the domain M of a smooth map- ping cj) : M —► N where AT is a smooth manifold endowed with a fixed Riemannian metric h so that cj) : (M, g) —► (AT, h) will be a harmonic mor- phism. The second result, Theorem 2, is a general finiteness theorem for har- monic morphisms of corank one with a given Riemannian domain (Mn+1, g) where n > 3. This result shows that the set of such harmonic morphisms is, in a certain sense, finite dimensional. This result is in marked contrast to the case n = 2, which has already been analyzed by Baird and Wood with the result that the locally defined harmonic morphisms with a given Rieman- nian domain (M3,g) of constant sectional curvature depend on arbitrary functions (in the sense of exterior differential systems). The third result, Theorem 3, is a classification of the harmonic mor- phisms of corank one whose domain (Mn+1,^) is a simply-connected, com- plete Riemannian manifold of constant curvature and dimension n+1 > 4. It will be shown that there are exactly two types of such harmonic morphisms. The first type can be thought of as a sort of metric quotient and is described as follows: Let X be a Killing vector field on M with zero lo- cus Z C M and suppose that the space N of integral curves of X in M\Z can Harmonic morphisms with fibers of dimension one 221 be given the structure of a smooth n-manifold in such a way that the quotient map (f): M\Z —>iVisa smooth submersion. Then there exists a metric h on N, unique up to a constant scale factor, so that </> : (M \Z,g) —» (iV, h) is a harmonic morphism. (Sometimes this map can be extended across Z as well after suitably extending AT, see §3.3.) The second type is described as follows: Let iV C M be a totally umbilic hypersurface, endowed with a constant multiple of the induced metric, denoted h. Let P C M be the focal set of AT, which consists of at most two points. There is a canonical retraction (j): M\P —> N that retracts M\P back to N along the geodesies normal to iV. Then ^ is a harmonic morphism. The examples of this kind had already appeared in the work of Gudmundsson [Gul]. The methods used are those of exterior differential systems and the mov- ing frame, both of which are well-adapted to the study of overdetermined systems of PDE. Acknowledgments. It is a pleasure to thank John Wood, whose questions inspired this article and whose comments and and guide to the literature on harmonic morphisms were invaluable. 1. Harmonic morphisms via moving frames. This section is a self-contained treatment by moving frame calculations of the the basic structure theory of harmonic morphisms. It is intended to be readable by those familiar with either moving frame calculations or the fundamentals of harmonic morphisms. Its main purpose is to fix notation and to serve as a reference for the proofs in the later sections, which employ the moving frame. Such a ref- erence is probably needed, as it appears that most of the current workers on harmonic morphisms do not use moving frames and so the translation of known results into this language may be helpful. For more background on the method of the moving frame, see [Sp]. Most of the results about harmonic maps and morphisms to be derived in this section can be found in the standard references on the subject, such as [ELI], [EL2], or [W2]. 1.1. Moving frame computations for harmonic morphisms. Let M and N be Riemannian manifolds of dimensions m and n, respectively. For simplicity, I assume that both M and N are connected throughout this article. The summation convention will be used extensively, with the 222 Robert Bryant understood ranges 1 < a, fe, c < m, 1 < iijjk < n. 1.1.1. Coframings, connection forms, and structure equations. Let g be the metric on M and h be the metric on N. Let U C M and V C N be open sets with trivial tangent bundles. Then there exist smooth cofram- ings u = (a;i,...,a;m) and 77 = (771,... ,ryn) on U and V respectively, so that h\v = r)i + -- + vl = Vi2- Corresponding to the chosen coframings on the respective open sets, there exist unique 1-forms u;a& = — c^ and rjij = —rjji that represent the Levi- Civita connections of the respective metrics and that are characterized by the structure equations (2) duja = -u>ab A ujb , drji = —rjij A rjj . 1.1.2. Mappings and pullbacks. Now suppose that (/> : M —» N is a smooth map and that U and V have been chosen so that U C /_1(y). Then there exist unique functions fia on £/ so that (3) <f)*(Vi) = fiaUa- Because the chosen coframings are orthonormal, the energy density of the map cj) on U is given by E(<t>)\U = /ia/ia |^1 A . AU;n| This density is globally defined, independent of the local choice of u or TJ. When M is compact, integration of this density yields a functional called the energy £ : C00(M, AT) -> R, namely Adopt the convention that, for any differential form ip on V, its <^- pullback </)*('0) on [/ is denoted by an overbar, i.e., ^ = (j)*(ip). Since Harmonic morphisms with fibers of dimension one 223 the map </> will be fixed in this discussion, this should cause no confusion. Thus, (3) becomes (S7) rfi = fia Va . (The reader of other sources on moving frame calculations should be aware that many authors simply drop the pullback notation entirely, writing (3) in the even simpler form rji = fia^a- This has caused considerable confusion in some cases, a confusion I hope to avoid.) Taking the exterior derivative of (3) and using the structure equations (2) yields (dfia - fib Vba + fja Wj) A ^a = 0. By Cartan's Lemma, there exist unique functions /^ = /^a on U so that (4) dfia = fib Uba - fja TJij + fiab ^b - The tension field offionU is the tensor field 7"(0) = fiaa ei0(f> (where (ei,..